Specification Mining on the User Interface Level (PDF)

Master's Thesis from the year 2012 in the subject Computer Science - Software, grade: 1,0, Technical University of Darmstadt (Secure Software Engineering Group), language: English, abstract: In this thesis we propose another approach targeted at web applications that use Ajax, HTML5 and other modern web technologies to achieve a look and feel that is only known by desktop applications. These web applications are also referred to as Rich Internet Application (RIA).

The user interface is a part of an application the same way its business logic is. This requires software engineers to test the user interface to verify its correctness to the same extend they test the rest of the application.But testing graphical user interfaces is a tedious task without standardized testing technologies engi-neers use for business logic verification like unit testing.To address this shortcoming, research groups applied the concept of model-based testing to user interface testing. In model-based testing an abstraction of the actual graphical user interface is used and test sequences are generated from the model. The model as a formal specification can also be checked by a model checker or pose as a test oracle to validate the correctness of the application.Ideally such a model, or specification, can be generated, or mined, by an automated process. This process is usually referred to as "specification mining" and relieves the test engineer of the task to keep the specification up to date. Many tools and approaches have been proposed to target different kinds of applications and properties that the mining tool infers.

At europe's biggest software company SAP, where this thesis was manufactured, many RIAsarebe-ing developed based on a user interface library called SAPUI5. This library helps developers to create uniform looking RIAs. To test their graphical user interface and ensure quality software, our approach supports this testing process by using a web crawler known from search engines to automatically infer a specification or model from these interfaces. We achieved this, by modifying the Ajax-capable crawler Crawljax by A. Mesbah and using a new, more efficient and modular state abstraction based on XPath expressions. Our CrawljaxXT also addresses many bugs and shortcomings of the original Crawljax and thus allows for reliable creation of complete user interface models without overly detailed state abstractions.Concluding we successfully evaluated our mined models and its testing capabilities with a modified version of the free model-based testing tool Graphwalker.